遥感图像拼接中改进的图像预处理算法研究

Research on Improved Image Preprocessing Algorithm in Remote Sensing Image Mosaicing

针对无人机遥感影像拼接中因图像噪声大、光照差异大、视野中景物存在畸变、图像不清晰等问题的影响,使得图像特征不明显、特征点提取数量少、特征点误匹配率高、拼接效果不理想甚至拼接错误等问题出现,论文提出了一种优化的遥感图像预处理算法用于无人机的遥感影像拼接。对于图像中可能存在畸变的情况,针对遥感图像的特点,提出了遥感图像畸变校正算法;且将无人机拼接最后阶段进行的匀色处理改进后应用于预处理阶段;并针对遥感图像经平滑去噪后图像细节不明显的问题,提出了基于图像锐化的平滑处理,即在图像平滑处理后使用自定义的锐化滤波器对平滑后的图像进行卷积,以增强遥感影像的细节。实验表明,论文所提出的遥感图像与处理算法,能够有效减小图像噪声、校正畸变、均衡光强、增强图像细节,使接下来的拼接工作更加高效准确,且经过验证,在后期进行特征提取时,采用论文所提预处理算法后提取到的特征点数目比起采取其他常规图像预处理算法后提取到的数目平均增加了约16%左右。

Due to the effects of large image noise,large differences in illumination,distortions in the field of vision,and unclear images in UAV remote sensing image mosaicing,the image features are not obvious,the number of feature points is small,and the feature point mismatch rate is high,the splicing effect is not ideal or even splicing errors. This paper proposes an optimized remote sensing image preprocessing algorithm for remote sensing image mosaic of drones. For the case of possible distortion in the image,a remote sensing image distortion correction algorithm is proposed for the characteristics of remote sensing images,the color processing in the final stage of UAV splice is improved and applied to the pre-processing stage,and the remote sensing image is smoothed. After the denoised image has no obvious details,a smoothing process based on image sharpening is proposed. After the image is smoothed,a sharpening filter is used to convolute the smoothed image to enhance the detail of the remote sensing image.Experiments show that the remote sensing image and processing algorithm proposed in this paper can effectively reduce image noise,correct distortion,balance light intensity,enhance image details,and make the next splicing work more efficient and accurate. And after verification,in the later feature extraction,the number of feature points extracted by using the pre-processing algorithm proposed in this paper increases by an average of about 16% compared to the number of feature points extracted after other conventional image pre-processing algorithms are adopted.